_id_f46969e5-91b7-41b8-995c-9701ab370543_size900.jpg?#)
Sparcent Speeds Up Multi-Vector Search by Turning Centroids into Sparse Indices
Researchers have introduced Sparcent, a technique that uses learned sparse retrieval to accelerate the query-centroid matching stage of multi-vector dense retrieval, achieving up to 4.6 times speed-up over PLAID with virtually no extra memory.
Researchers have introduced Sparcent, a technique that uses learned sparse retrieval to accelerate the query-centroid matching stage of multi-vector dense retrieval, achieving up to 4.6 times speed-up over PLAID with virtually no extra memory.
What's Your Reaction?
